#!/usr/bin/env bash # Pull exactly what the 32k eval needs. Explicit --include on every call: # an unfiltered download of this repo once pulled 167 GB and broke a build. set -uo pipefail cd "$(dirname "$0")" export PATH=/root/Adhi/BluTrain/envv/bin:$PATH say(){ echo "[$(date -u +%H:%M:%S)] $*"; } mkdir -p checkpoints_680m/rank_0 checkpoints_bluscriptcp/rank_0 Data_Loader/Data/b2_8192_16384 Data_Loader/Data/evalsrc say "1/4 varybatch.ckpt (the untrained base -- the control)" hf download navingv/varybatch --quiet --include 'varybatch.ckpt' --local-dir dl_vb || exit 1 mv -f dl_vb/varybatch.ckpt checkpoints_680m/rank_0/ && rm -rf dl_vb say "2/4 p1final step1200 rank0 (5.14 GB)" hf download unparallelled/BluScriptCP --quiet --include 'checkpoints/p1_680m_step1200_rank0.ckpt' --local-dir dl_p1 || exit 1 mv -f dl_p1/checkpoints/p1_680m_step1200_rank0.ckpt \ checkpoints_bluscriptcp/rank_0/blumodelcp_run12_step_1200.ckpt && rm -rf dl_p1 say "3/4 haystack -- MUST be the same file phase 1 used or the numbers are not comparable" hf download unparallelled/BluScriptCP-data --repo-type dataset --quiet \ --include 'cp_data_by_length/b2_8192_16384/val_shard_00000.bin' --local-dir dl_hay || exit 1 mv -f dl_hay/cp_data_by_length/b2_8192_16384/val_shard_00000.bin \ Data_Loader/Data/b2_8192_16384/ && rm -rf dl_hay say "4/4 a train split for the loader -- the trainer builds one even for pure generation" hf download unparallelled/BluScriptCP-data --repo-type dataset --quiet \ --include 'cp_data_by_length/b1_4096_8192/train_shard_00000.bin' --local-dir dl_tr || exit 1 mv -f dl_tr/cp_data_by_length/b1_4096_8192/train_shard_00000.bin Data_Loader/Data/evalsrc/ && rm -rf dl_tr cp -f Data_Loader/Data/b2_8192_16384/val_shard_00000.bin Data_Loader/Data/evalsrc/ say "=== sizes ===" stat -c ' %s %n' checkpoints_680m/rank_0/varybatch.ckpt \ checkpoints_bluscriptcp/rank_0/blumodelcp_run12_step_1200.ckpt \ Data_Loader/Data/b2_8192_16384/val_shard_00000.bin \ Data_Loader/Data/evalsrc/train_shard_00000.bin 2>/dev/null say "expected: varybatch 1362674564 | p1final 5141244514 | haystack 199754694" say "DOWNLOAD DONE"